Research map / 3D vision / Gaussian splatting
Learning Differentiable Hierarchies in 3D Gaussian Splatting
Youqi Pan, Wugen Zhou, Hongbin Zha
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add hierarchy balance loss (ablation) | DDSF hierarchy learning · Truck sequence (avg over splat ratios) | PSNR: 21.76 -> 25.24 | Improved |
| use PDF-guided active-region sampling instead of random sampling (ablation) | DDSF hierarchy learning · Truck sequence (avg over splat ratios) | PSNR: 24.65 -> 25.24 | Improved |
| use PDF-guided sampling instead of quantile sampling (ablation) | DDSF hierarchy learning · Truck sequence (avg over splat ratios) | PSNR: 24.23 -> 25.24 | Improved |
| set sigmoid slope beta to 0.1 instead of 10 (ablation) | DDSF · Truck sequence (avg over splat ratios) | PSNR: 25.24 -> 22.49 | Hurt |
| set sigmoid slope beta to 1.0 instead of 10 (ablation) | DDSF · Truck sequence (avg over splat ratios) | PSNR: 25.24 -> 24.86 | Hurt |
| set sigmoid slope beta to 100 instead of 10 (ablation) | DDSF · Truck sequence (avg over splat ratios) | PSNR: 25.24 -> 24.11 | Hurt |
| use single-stage learned hierarchy instead of CLOD two-stage training | LoD rendering · Mip-NeRF360, 100% splats | PSNR: 29.59 -> 29.74 | Improved |
| use single-stage learned hierarchy instead of CLOD | LoD rendering · Mip-NeRF360 | Train time (min:s): 59:34 -> 26:41 | Improved |
| add hierarchy learning to 3DGS-MCMC | 3DGS-MCMC · Mip-NeRF360 | Train time (min:s): 23:53 -> 26:41 | Hurt |
| use dedicated pruning training instead of 60% LoD model (ablation) | Ours on 3DGS-MCMC · Mip-NeRF360 pruning | PSNR: 29.59 -> 29.67 | Improved |
| use learned hierarchy instead of CLOD at 50% splats on Mip-NeRF360 | LoD rendering · Mip-NeRF360, 50% splats | PSNR: 29.51 -> 29.47 | No clear effect |
Closest papers by meaning
- LODGE: Level-of-Detail Large-Scale Gaussian Splatting with Efficient Rendering (NeurIPS 2025 · similarity 0.92)
- LOD-GS: Achieving Levels of Detail using Scalable Gaussian Soup (CVPR 2025 · similarity 0.92)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.91)
- Lightweight Predictive 3D Gaussian Splats (ICLR 2025 · similarity 0.91)
- HiCoM: Hierarchical Coherent Motion for Dynamic Streamable Scenes with 3D Gaussian Splatting (NeurIPS 2024 · similarity 0.91)
- H3D-DGS: Exploring Heterogeneous 3D Motion Representation for Deformable 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
